PTDC/EIA-EIA/119004/2010 - Reusable Deep Neural Networks: Applications to Biomedical Data (ongoing)

Deep architectures, such as neural networks with two or more hidden layers of units, are a class of machines that comprise several levels of non-linear operations, each expressed in terms of parameters that can be learned. In this project we investigate various aspects of deep networks, such as their training via the use of different cost functions, their reusability, and their application to the analysis of biomedical data. We also aim to use larger datasets using GPU processing.

This project is financed by FEDER funds through the Programa Operacional Factores de Competitividade COMPETE and by Portuguese funds through FCT Fundação para a Ciência e a Tecnologia.

Team Members:

Joaquim P. Marques de Sá, University of Porto

Jorge M. Santos, School of Engineering, Polytechnic Institute of Porto

Luís A. Alexandre, University of Beira Interior

Luís M. Silva, University of Aveiro

Ricardo Sousa, Post-Doctoral Investigator

Chetak Kandaswamy, Research Assistant

Telmo Amaral, Post-Doctoral Investigator (Former)

 Deep Transfer Learning Software Interface

 [Software]

 Publications

 Journal:

1. Fontes, T., Luís M. Silva, M. P. Silva, N. Barros, and A. C. Carvalho. "Can artificial neural networks be used to predict the origin of ozone episodes?." Science of the Total Environment 488 (2014): 197-207. DOI: 10.1016/j.scitotenv.2014.04.077 [pdf]
   
2. Sousa,  Ricardo Gamelas, Joaquim Marques de Sá, Luis A. Alexandre, Jorge M. Santos, Luis M. Silva. "Classifier Transfer Learning: A Survey Towards A Unifying View". (submitted)
   
3. Kandaswamy, Chetak, Luís M. Silva, Luís A. Alexandre, and Jorge M. Santos. "High-content Analysis of Breast Cancer using Single-Cell Deep Transfer Learning". (submitted) [Software] [Code] [Data] 
   
4. "Stacked Denoising Autoencoders and Transfer Learning for Immunogold Particles Detection and Recognition" (submitted) [supplementary information] [Data]
   

Conferences:

1. Sofia Fernandes, Ricardo Sousa, Renato Socodato and Luís M. Silva. "Stacked Denoising Autoencoders for the Automatic Recognition of Microglial Cells' State". In proceedings of ESANN, April 2016, Brueges, Belgium. [Additional info]
2. Sousa. Ricardo Gamelas, Tiago Esteves, Sara Rocha, Francisco Figueiredo, Pedro Quelhas, and Luís M. Silva. "Automatic Detection of Immunogold Particles from Electron Microscopy Images". In proceedings, ICIAR, July 22-24, 2015 – Niagara Falls, Canada, (Accepted)
   
3. Sousa. Ricardo Gamelas, Tiago Esteves, Sara Rocha, Francisco Figueiredo, Joaquim M. de Sá, Luís A. Alexandre, Jorge M. Santos, and Luis M. Silva. "Transfer Learning for the Recognition of Immunogold Particles in TEM imaging."  In Advances in Computational Intelligence, IWANN. Springer International Publishing, 2015. DOI: 10.1007/978-3-319-19258-1_32 [pdf].
4. Kandaswamy, Chetak, Luís M. Silva, Luís A. Alexandre, and Jorge M. Santos. "Deep Transfer Learning Ensemble for Classification." In Advances in Computational Intelligence, IWANN, pp. 335-348. Springer International Publishing, 2015. DOI: 10.1007/978-3-319-19258-1_29 [pdf].
5. Kandaswamy, Chetak, Luís M. Silva, and Jaime S. Cardoso. "Source-target-source classification using Stacked Denoising Autoencoders." In Pattern Recognition and Image Analysis, IbPRIA, pp. 39-47. Springer International Publishing, 2015. DOI: 10.1007/978-3-319-19390-8_5 [pdf]
6. Amaral, Telmo, Luís M. Silva, Luís A. Alexandre, Kandaswamy. Chetak, Joaquim Marques de Sá, and Jorge M. Santos. "Transfer learning using rotated image data to improve deep neural network performance." In Image Analysis and Recognition, ICIAR, pp. 290-300. Springer International Publishing, 2014. DOI: 10.1007/978-3-319-11758-4_32 [pdf]
7. Kandaswamy, Chetak, Luís M. Silva, Luis Alexandre, Ricardo Sousa, Jorge M. Santos, and Joaquim Marques de Sá. "Improving transfer learning accuracy by reusing stacked denoising autoencoders." In Systems, Man and Cybernetics (SMC), October 5-8, San Diego, CA, USA, 2014. IEEE International Conference on, pp. 1380-1387. DOI: 10.1109/SMC.2014.6974107. [pdf]
8. Amaral, Telmo, Chetak Kandaswamy, Luís M. Silva, Luis Alexandre, Joaquim Marques De Sa, and Jorge M. Santos. "Improving performance on problems with few labelled data by reusing stacked auto-encoders." In Machine Learning and Applications (ICMLA), 2014 13th International Conference on, pp. 367-372. IEEE, , Detroit, USA, December 3-6, 2014. DOI: 10.1109/ICMLA.2014.65. [pdf]
   
9. Alexandre, Luís A. "3D object recognition using convolutional neural networks with transfer learning between input channels." In Proc. the 13th International Conference on Intelligent Autonomous Systems. Springer, July 15-18, Padova, Italy, 2014. [pdf]
10. Kandaswamy, Chetak, Luís M. Silva, Luís A. Alexandre, Jorge M. Santos, and Joaquim Marques de Sá. "Improving deep neural network performance by reusing features trained with transductive transference." In Artificial Neural Networks and Machine Learning–ICANN 2014, pp. 265-272. Springer International Publishing, 2014. .DOI:10.1007/978-3-319-11179-7_34. [pdf] 
11. Amaral, Telmo, Luís M. Silva, Luís Alexandre, Kandaswamy. Chetak, Jorge M. Santos, and Joaquim Marques de Sá. "Using different cost functions to train stacked auto-encoders." In Artificial Intelligence (MICAI), 2013 12th Mexican International Conference on, pp. 114-120. IEEE,  November 24-30, 2013, Mexico City, Mexico, 2013. DOI: 10.1109/MICAI.2013.20. [pdf]
12. Fontes, Tânia, Luís M. Silva, Sérgio R. Pereira, and Margarida C. Coelho. "Application of artificial neural networks to predict the impact of traffic emissions on human health." In Progress in Artificial Intelligence, pp. 21-29. Springer Berlin Heidelberg,Angra do Heroísmo, Açores, Setembro de 2013. DOI: 10.1007/978-3-642-40669-0_3.  [pdf].
13. Kandaswamy, Chetak, Luís M. Silva, Jaime S Cardoso. "Improving Classification Accuracy of Deep Neural Networks by Transferring Features from a Different Distribution", 20th edition of the Portuguese Conference on Pattern Recognition, University of Beira Interior, Covilhã, 2014. [pdf].
14. Sousa, Ricardo, Luis M. Silva, Luis A. Alexandre, Jorge Santos, and Joaquim Marques de Sá. "Transfer Learning: Current Status, Trends and Challenges." 20th edition of the Portuguese Conference on Pattern Recognition, University of Beira Interior, Covilhã, 2014. [pdf].

Technical Reports:

1. C. Kandaswamy. Improve Performance in Deep Neural Networks: (1) Cost Functions, and (2) Reusable learning, Feb 2014, NNIG Technical Report No. 3/2014.[pdf]

2. C. Kandaswamy, L. M. Silva, L. A. Alexandre. Report: Improving CNN by Reusing Features Trained with Transductive Transfer Setting , Feb 2014, NNIG Technical Report No. 2/2014. [pdf]

3. T. Amaral . Transfer of Learning Across Deep Networks to Improve Performance in Problems with Few Labelled Data , Jan 2014 . NNIG Technical Report No. 1/2014 . [pdf]

4. T. Amaral . Using Different Cost Functions to Train Deep Networks with Supervision , Jul 2013, NNIG Technical Report No. 3/2013 . [pdf]

5. C. Kandaswamy, T. Amaral. Tuning Parameters of Deep Neural Network Algorithm for identifying best Cost function.  NNIG-INEB Technical Report 2/2013. [pdf]

6. T. Amaral, L. M. Silva, L. A. Alexandre. Using different cost functions when pre-training stacked auto-encoders.  NNIG-INEB Technical Report 1/2013. [pdf]

7. T. Amaral.  Experiments with a restricted Boltzmann machine. NNIG-INEB Technical Report 1/2012. [pdf] 

Activities Reports:

• Sousa, Ricardo Gamelas. Final Report (Work carried out from March 2014 to February 2015). [pdf]
• Kandaswamy, Chetak. Activities report from March 2014 to March 2015. [pdf]
• Kandaswamy, Chetak. Activities report from March 2013 to March 2014. [pdf]
• Amaral, Telmo. Overview of Archived Materials , Jan 2014. [pdf]
• Amaral, Telmo. Activities report from August 2012 to April 2013. [pdf]
  

Upcoming Events:

We are organizing a special session on Transfer Learning at IWANN conference at Spain on 15th September 2015.

Code for repoducing MFC7 breast cancer experiments using DTL:

DTL transfer learning code: https://github.com/chetakks/DTL

Download Dataset: ljosa_data